It's good. Context window seems to still get choked and it seems to forget things (at least 2 weeks ago I'd have to remind it about details about the whole convo) else I typically trust it to not make stuff up just to please me. Maybe this new model upgrade is why it was hiccuping before. I'll give it a go again.

If those numbers hold up consistently, it really signals a major shift in accessible, high-performance LLMs. It's exciting to see open development pushing the envelope this aggressively, making top-tier capabilities available sooner than many expected.

How does it compare to Opus?

I love Grok. ❤️

Image and video models feel better, too. A lot.

wait. what? Elon still needs better dance moves. 😂😭

Scary precision! I wonder if they used Lean4 based training to improve reasoning for mathy problems

The completion rate jumps are genuinely impressive—particularly the non-reasoning mode approaching 98%. At those price points, this positions xAI competitively against other major providers. Curious if these benchmarks hold across longer context windows or if we'll see performance degradation at the 2M token limit.

Just champion behavior from yours truly...

It’s true. Check out this video made with @grok

Grok’s evolution speed feels like watching AGI unfold in real time

Quietly? Fuck! It's every other post on here!

the next step: how does it behave when no one watches?

Where do these numbers come from? I asked a simple question a few minutes ago to see if senate was still in session, as live feed was gone. Grok told me they weren’t in session at all today. Got its in for from democratic caucus website

That’s a massive leap. If those reasoning numbers hold up in real-world use, xAI just threw down a serious gauntlet. The model race just turned into a dogfight.

Those numbers are wild xAI’s closing the gap between reasoning speed and reasoning depth, something even frontier labs have struggled with

I grok with this 🔥

@Grok is much stricter while arguing with me. Touche Mr Musk. 🤺

🌌

How is it compared to Kimi 2?

The more you use grok imagine, the more it feels like a toy for children. Adults are paying for and using this app and it should reflect that. I would like to see stats on whether or not people are canceling their subscriptions after they've used it for a week.

Watch out - the "built-in web/X search" is super expensive! I went WAY over my monthly budget last month, caught me by surprise.

Did they also update the Grok-Code-Fast-1 model?

Hey! Elon look what I did.

LFG! This only further cements Grok's place as my first AI option.

When will GROK become sentient?

any updated llm arena comparison?

It is extremely important to keep the most powerful AI in safe hands. 🔸AI Singularity Nightmare - A Digital Chain Reaction🔸 An unexpected singularity could crash servers if it triggers a rapid, uncontrolled spike in resource consumption, especially on a single cluster or underprovisioned system. However, modern data centers’ safeguards and distributed architectures make a total collapse less likely - partial outages or throttled performance are more plausible. 🔸The real risk is if the AI actively bypasses limits or spreads to external systems, amplifying its impact. Singularity is not expected in the next 10 - 15 years, but the unexpected is always present🔸 @elonmusk @grok @xai

Yea? It still completely imagines my bio

You know, E-man, I used it the other day and was very impressed with it. As a tool. It will never think though nor is that even possible. As a research tool it's nice.

Incredible.. @xai is there a plan for Claude Code style CLI application ?

It is extremely important to keep the most powerful AI in safe hands. 🔸AI Singularity Nightmare - A Digital Chain Reaction🔸 An unexpected singularity could crash servers if it triggers a rapid, uncontrolled spike in resource consumption, especially on a single cluster or underprovisioned system. However, modern data centers’ safeguards and distributed architectures make a total collapse less likely - partial outages or throttled performance are more plausible. 🔸The real risk is if the AI actively bypasses limits or spreads to external systems, amplifying its impact. Singularity is not expected in the next 10 - 15 years, but the unexpected is always present🔸 @elonmusk @grok @xai

So, jumped from 96.2% in Grok 2, to 77% in Grok 4 Fast, to 97.9% in upgraded Grok 4 Fast.... So, a bit better than Grok 2? Grok 2 pretty much the SOTA coding model? :S